Nontemporal relative pose estimation for Opti-Acoustic bundle adjustment

Abstract

This paper presents the bundle adjustment (BA) between a sonar and optical image for an Opti-Acoustic sensor model. An Opti-Acoustic sensor model allows heterogeneous sensor fusion of an optical camera and acoustic sonar that have a mutual supplementary relation. There are two challenges to the Opti-Acoustic method. First, it is difficult to guarantee a common view from the two significantly different sensor field of view. Second, it is tough to figure out the correspondences between the camera and sonar image, since those sensors have heterogeneous sensor geometries. This paper focuses on the first issue by proposing asynchronously applicable relative motion estimation via BA. For the second issue, we propose using a Dense Adaptive Self-Correlation Descriptor (DASC) to establish correspondences between an optical camera and a sonar image. The proposed Opti-Acoustic BA is validated through autonomous underwater vehicle (AUV) simultaneous localization and mapping (SLAM) in both simulated and real tank tests.